From cebf5f267f5761f3cbec094ce5ffa1c377f1d761 Mon Sep 17 00:00:00 2001
From: Nikolai <osterei33@gmx.de>
Date: Fri, 27 Jul 2018 10:14:06 +0200
Subject: [PATCH] switch to np.digitize for histogram error calculation
---
toolkit.py | 20 +++++++++++++-------
1 file changed, 13 insertions(+), 7 deletions(-)
diff --git a/toolkit.py b/toolkit.py
index 9a0a445..8d36861 100755
--- a/toolkit.py
+++ b/toolkit.py
@@ -828,13 +828,19 @@ class ClassificationProject(object):
hist, bins = np.histogram(x, **np_kwargs)
centers = (bins[:-1] + bins[1:]) / 2
if "weights" in np_kwargs:
- errors = []
- for left, right in zip(bins, bins[1:]):
- indices = np.where((x >= left) & (x < right))[0]
- sumw2 = np.sum(np_kwargs["weights"][indices]**2)
- content = np.sum(np_kwargs["weights"][indices])
- errors.append(math.sqrt(sumw2)/content)
- errors = np.array(errors)
+ bin_indices = np.digitize(x, bins)
+ sumw2 = np.array([np.sum(np_kwargs["weights"][bin_indices==i]**2)
+ for i in range(1, len(bins)+1)])
+ sumw = np.array([np.sum(np_kwargs["weights"][bin_indices==i])
+ for i in range(1, len(bins)+1)])
+ # move overflow to last bin
+ # (since thats what np.histogram gives us)
+ sumw2[-2] += sumw2[-1]
+ sumw2 = sumw2[:-1]
+ sumw[-2] += sumw[-1]
+ sumw = sumw[:-1]
+ # calculate relative error
+ errors = np.sqrt(sumw2)/sumw
else:
errors = np.sqrt(hist)/hist
if scale_factor is not None:
--
GitLab